DEI/CISUC Seminars

Publication Date: 2018-12-17 12:06:09

December 19, Wednesday,
16h (sharp),

Invited Speaker: Dimo Brockhoff

Title: Benchmarking multiobjective optimizers: An algorithmic jam session of recent results

Abstract: Blackbox optimization problems appear in many application areas and, consequently, many blackbox (or derivative-free) optimizers are available these days. One of the main practical questions, the choice of an appropriate algorithm for a concrete problem at hand, is approached best by a data-driven approach in which available algorithms are benchmarked and the algorithm with the best expected or forecast performance on the problem at hand is chosen.  This scenario is not different, in general, if multiple objective functions must be optimized simultaneously. But the benchmarking methodology and the understanding and availability of basic artificial test functions is arguably less developed in this multiobjective scenario compared to the single-objective case.In this talk, I will present a mix of recent results around benchmarking multiobjective optimizers and discuss in particular:

- our extension of attainment function plots towards displaying average runtimes,
- our biobjective blackbox optimization benchmarking framework and the COCO software for automated benchmarking, and
- our latest result on the properties of multiobjective convex quadratic functions.

Wherever possible, I will highlight open questions and potential extensions that can serve as starting points for discussions after the talk.

Short bio: Dimo Brockhoff received his diploma in computer science from University of Dortmund, Germany in 2005 and his PhD (Dr. sc. ETH) from ETH Zurich, Switzerland in 2009. Afterwards, he held two postdoctoral research positions in France at Inria Saclay Ile-de-France (2009-2010) and at Ecole Polytechnique (2010-2011) before joining Inria in November 2011 as a permanent researcher (first in its Lille - Nord Europe research center and since October 2016 in the Saclay - Ile-de-France center). His research interests are focused on evolutionary multiobjective optimization (EMO), in particular on theoretical aspects of indicator-based search and on the benchmarking of blackbox algorithms in general.